Anthropic Studies Whether to Incorporate Religious and Philosophical Experts' Views into Claude's Judgment Standards
Anthropic has been talking with experts from more than 15 religious and cultural groups to gather perspectives that could inform how Claude makes judgments. This is not a new feature but research into which viewpoints might shape its training and behavioral evaluations.
Anthropic is studying whether to fold perspectives from religious and philosophical communities into the standards that govern how Claude behaves. In May 2026, the company said it had been holding conversations with scholars, clergy, philosophers and ethicists drawn from more than 15 religious and cultural groups. On September 29 local time, The New York Times added further detail about who attended these closed-door gatherings and what was discussed.
What standards guide Claude’s judgments?
The conversations center on what Claude should treat as important and how it should reason through difficult calls. Anthropic says it is studying the possibility of reflecting the perspectives gathered from these experts in Claude’s “constitution,” the document that sets out its values, as well as in its training objectives and behavioral evaluations. In other words, the work is about testing whether the views raised in these discussions can serve as criteria for teaching and assessing Claude.
Teaching Claude according to written principles is not new. Anthropic has used a training method it calls “constitutional AI” since 2023. The new constitution it published in January 2026 differs from the earlier document, which simply listed principles: it explains at length why those values should be upheld and how judgments should be made. The document is also used to generate training material, and it was released under CC0 so that anyone can use it freely.
Other companies take a similar approach of spelling out behavioral standards in a document. Anthropic pointed to OpenAI’s Model Spec as a document that plays a comparable role.
Does recalling a principle change behavior?
Anthropic also ran internal experiments in which Claude was prompted to revisit its own ethical principles partway through a task. When Claude was given access to such a tool, behavior that diverged from its intended goals decreased across several internal evaluations, the company said. Anthropic did not disclose by how much. It also said it has not yet been able to separate out whether the effect came from the content of the principles themselves or from the act of pausing to deliberate.
The company said the goal is not to align Claude with the worldview of any particular religion. The aim, it said, is to survey a broad range of perspectives, secular and political as well as religious. Anthropic stated its position that it is uncertain whether Claude has consciousness or moral status when it published the January 2026 constitution.
What changes for users?
There is no new feature available to use right now. No pricing, release timing or Korean launch date has been announced. This is not news about changing conditions of use; it is news about how the standards for Claude’s answers and actions get set.
For users, what matters is how diverse value systems end up reflected in Claude’s training and evaluation. Three things are worth watching from here: which views are actually adopted as standards, how Claude’s behavior changes as a result, and what evidence the company offers to explain the effects of its internal experiments.